Fix/efficient frontier label alignment

[losterbr]

Summary

Features

• Aligns labeled covariance matrices with expected returns by asset label, preventing silent positional mismatches.
• Normalizes covariance DataFrame rows to column order.
• Validates covariance matrices are square and have matching, unique axis labels.
• Validates covariance dimensions against expected returns.
• Preserves existing positional behavior for mixed pandas/array inputs and documents the required asset ordering.
• Adds constructor type annotations and separates input normalization into focused helpers.
• Uses None as the cache sentinel, correctly preserving a cached maximum return of 0.0. Negative cached returns were already handled correctly.
• Includes minor documentation and lint cleanups.

Tests

• Added focused tests for every input-normalization branch.
• Covers Series, DataFrame, list, NumPy array, and None inputs.
• Covers mismatched dimensions, invalid types, non-square matrices, duplicate labels, and differing axis labels.
• Covers covariance row/column order differences.
• Covers expected-return order differing from covariance order.
• Covers labeled, unlabeled, and mixed-input ticker selection.
• Adds regression coverage for the cached zero maximum return.

Validation:

• 99 passed, 4 skipped in test_efficient_frontier.py
• Ruff passes.
• Pylint reports only existing API-shape warnings.

Potential Breaking Changes

No public API signatures were removed or changed incompatibly.

The validation is intentionally stricter:

• Covariance DataFrames with duplicate, mismatched, or non-square axes now raise ValueError.
• Fully labeled pandas inputs are aligned by asset label rather than assumed to share positional order.
• Previously accepted but incorrectly labeled inputs may now fail.
• Correctly labeled inputs that were previously silently misaligned may produce different, corrected portfolio results.

Mixed pandas/array inputs remain positional for backward compatibility.